none
DPM2010 and Enabling Per CSV LUN Serialization RRS feed

  • Question

  • Hi,

    My SAN currently does not have enabled hardware VSS provider.
    So my question is, if I enabled per CSV LUN serialization ( created required xml ) do I have to create following registry setting too?

    "HKLM\Software\Microsoft\Microsoft Data Protection Manager\2.0\Configuration\MaxAllowedParallelBackups\Microsoft Hyper-V=1"

    Or using this xml based serialization makes this key irrelevant?


    rgds Sven
    Sunday, January 29, 2012 1:23 PM

Answers

  • Hi,

    It is recommended that you also set the MaxAllowedParallelBackups to one.  Since we're using software snapshots and you may have multiple CSV's it is then possible for more than one guest to be backed up (one from each CSV) on the same host and effect performance due to the software snapshots being maintained during backup.  Setting the MaxAllowedParallelBackups to 1 will prevent that. If you believe your system is beefy enought to handle that, you can try setting it to 2 and see what happens, but most customers leave it at 1.  Let me know your experience if you decide to try 2 or more.


    Regards, Mike J. [MSFT] This posting is provided "AS IS" with no warranties, and confers no rights.
    Sunday, January 29, 2012 5:39 PM
    Moderator

All replies

  • Hi,

    It is recommended that you also set the MaxAllowedParallelBackups to one.  Since we're using software snapshots and you may have multiple CSV's it is then possible for more than one guest to be backed up (one from each CSV) on the same host and effect performance due to the software snapshots being maintained during backup.  Setting the MaxAllowedParallelBackups to 1 will prevent that. If you believe your system is beefy enought to handle that, you can try setting it to 2 and see what happens, but most customers leave it at 1.  Let me know your experience if you decide to try 2 or more.


    Regards, Mike J. [MSFT] This posting is provided "AS IS" with no warranties, and confers no rights.
    Sunday, January 29, 2012 5:39 PM
    Moderator
  • Okay, but how should I understand the technet article ( http://technet.microsoft.com/en-us/library/ff634192.aspx ) wich says, to use one or another way :
    1. Serializing virtual machine backups on a per node basis
    2. Serializing virtual machine backups on a per CSV LUN basis

    There's no statement to use both of them. If I understand correctly, then using xml + reg key method is the most foolproof option but also most slowest?

    Also, if I have two CSV's (fast and slow) and two node cluster. All guest are using VHD's only on CSV. One guest_A (using fast disks) are on Node1 and the other guest_B (using fast disk) are on Node2 + bunch of guest (using slow disk) on both nodes.
    Then what happend when my DPM2010 is configured to backup cluster using child nodes? My xml contains TWO groups (because I have two CSV) and my registry is set up to backup one-at-the time AND I have created ONE protection group for all child nodes. What the backup logic will be? My mind is bit messed up what would be the best serializing logic on this situation?
    Creating TWO protection groups? One group for fast CSV and another for slow.. or create TWO protection group based on nodes? Or something else?

    BTW. Is there a way to configure DPM protection groups to dependent eachother, like if the first group finishes then the next one starts?

     


    rgds Sven
    Sunday, January 29, 2012 6:46 PM
  • Hi,

    <Snip>
    Q) My xml contains TWO groups (because I have two CSV) and my registry is set up to backup one-at-the time AND I have created ONE protection group for all child nodes. What the backup logic will be?
    >Snip<

    A) DPM will only backup two guests simultaniously, one guest on CSV1 on NodeA and one guest on CSV2 on NodeB.  If you have an uneven distribution of guests between the two CSV's, then the backups will take longer since we can only backup one guest per CSV at a time when using the software provider.

    DPM Moves the CSV disk to the host that is running the guest when it's time to backup that guest.  If you want to minimize moving the CSV between nodes, then ideally you would have all guests locaed on CSV1 running on NodeA and all guests located on CSV2 on NodeB when it comes time to backup.  It doesn't matter if you have one PG or multiple PG's, DPM is going to control how many guests are backed up at one time based on the serialization.

    Of course we al know the purpose of a cluster and you can't always control what guest is running on what host when it's time to backup, so bottom line it doesn't matter, as long as serialization is setup and maintained as you add more guests, the backups should succeed.  If you find the backups are taking longer that your desired backup window, you can either enable the VSS hardware provider so more simultanious backups can  be done per node, or add more nodes and CSV disks and distribute the guests across each.

     


    Regards, Mike J. [MSFT] This posting is provided "AS IS" with no warranties, and confers no rights.
    Monday, January 30, 2012 1:53 AM
    Moderator
  • Thank You for Your time. This information is useful for me :)
    rgds Sven
    Monday, January 30, 2012 8:17 AM